The Path to Analytics Maturity—Step 2: Applying Specific Analytics
This is the second post in our four-part weekly series exploring analytics maturity. If you missed the first post in the series, catch up by reading Step 1: Monitoring Critical Assets.
Once a customer has mastered basic monitoring, they often want to begin their analytics journey by deploying very specific analytics to improve operations. They typically begin by using analytics to analyze a problem. Maybe a piece of equipment alarms and needs debugging, so the team uses basic analytics to troubleshoot the problem. While identifying the cause, they’ll often want to create an analytic to avoid the problem altogether.
A classic example here is PID loops. A customer wants to drive constant output given variability in input and creates PID loops to optimize for known variation, stabilizing process variability. Most customers in most industries have used and leveraged the value of PID loops to save millions of dollars by avoiding the unneeded wear and tear on equipment that results from large-scale process variation and variability. Tying this to their monitoring stage is the next “best practice.” One way to do this is to create confidence intervals on PID loops so you can understand how well the PID loops are behaving, given process variation. Of course, new equipment, new materials, new processes and new operations can create new input variation that PID loops are not tuned for. By creating confidence intervals on PID loops, customers are able to monitor the confidence interval as a KPI. And when input variation occurs that drives unplanned changes in system behavior, they are able to see that their confidence interval decreases, alerting them to the need to tune their PID loops to the new input realities.
One customer was able to monitor their PID loops via a remote GE service and identified that many PID loops were running in manual mode because they needed tuning (the operators had shut them off to try to manually control the process instability). Turning off PID loops is common, but many customers also forget to turn them back on, or don't understand that they are running in manual mode. Because they monitored the PID loops, one customer was able to save $2 million by retuning their loops. Without the monitoring, they would not have known they needed the service.
There are, of course, many examples of using analytics to solve very specific needs. In fact, many monitoring sites can build up hundreds—even thousands—of specific analytics that look for very specialized problems. We have one customer who runs these thousands of times a day, resulting in millions of analytics being executed every day! This behavior is typical of an organization that is new to analytics. They have a strong belief in their own engineering skills to identify problems, troubleshoot issues and build specific solutions to those problems. At this stage, however, most customers have not taken the leap that machine learning analytics can be better at finding and avoiding problems than their engineers.